Novel CBIR System using CNN Architecture

Development of multi-media technologies large number of images are used in various fields such as video satellite data, medical treatment and digital judicial systems and surveillance systems. An efficient representation of features from an image for retrieval process is a challenging task. In this paper provides the feature extraction of an image using deep learning technique to tackle the differences between low-level features and high-level semantic features of basic CBIR systems. In this technique feature database can be created from each image in the database using VGG 16 model. By using Euclidean distance metrics an image analogous to the image of the query was retrieved by comparing the feature vector of the query image (compute similar to the data base images) and the feature database. The results suggest that the proposed CNN techniques yields better results than the other existed techniques.